最近開始嘗試深度強(qiáng)化學(xué)習(xí)，很好的一個(gè)環(huán)境平臺(tái)是Gym: https://www.gymlibrary.dev/content/basic_usage/

安裝： pip install gym[all]

強(qiáng)化學(xué)習(xí)、深度強(qiáng)化學(xué)習(xí)資源：

• 教程：
  1. 深度強(qiáng)化學(xué)習(xí)_王樹森 https://www.bilibili.com/video/BV1rv41167yx
  2. 莫煩PYTHON https://mofanpy.com/tutorials/machine-learning/reinforcement-learning/intro-RL
• 方向
  1. 自動(dòng)駕駛模擬仿真 Carla
• 項(xiàng)目
  1. 強(qiáng)化學(xué)習(xí)git項(xiàng)目總結(jié) https://zhuanlan.zhihu.com/p/24392239
  2. 強(qiáng)化學(xué)習(xí)開源項(xiàng)目 https://www.zhihu.com/question/49230922/answer/1631248683
• 論文
  1. OpenAI Gym
  2. Multi-Agent Path Finding
  3. RL最新論文
• 其他資源
  • DeepMind
  • OpenAI
  • CS285
  • paper with code
  • Spinning up

Jupyter 中顯示gym渲染窗口及保存為gif

參考:

• https://cloud.tencent.com/developer/article/1879482

• https://www.cnblogs.com/cenariusxz/p/12666938.html

• https://blog.csdn.net/qq_33361420/article/details/112471755?spm=1001.2014.3001.5502

• https://zhuanlan.zhihu.com/p/527274837

import numpy as np
import time 
import gym
import matplotlib.pyplot as plt 
from matplotlib import animation 
%matplotlib inline
from IPython import display

# 顯示gym渲染窗口的函數(shù)，在運(yùn)行過程中將 env.render() 替換為 show_state(env, step, info).
def show_state(env, step=0, info=""):
    plt.figure(3)
    plt.clf()
    plt.imshow(env.render(mode='rgb_array'))
    plt.title("Step: %d %s" % (step, info))
    plt.axis('off')

    display.clear_output(wait=True)
    display.display(plt.gcf())

def display_frames_as_gif(frames, SavePath = './test.gif'):
    patch = plt.imshow(frames[0])
    plt.axis('off')
    def animate(i):
        patch.set_data(frames[I])

    anim = animation.FuncAnimation(plt.gcf(), animate, frames = len(frames), interval=1)
    anim.save(SavePath, writer='ffmpeg', fps=30)

# 運(yùn)行環(huán)境實(shí)例1

import gym

frames=[]
env = gym.make('CartPole-v1')
info = env.reset() # 重置環(huán)境 

for step in range(100):
    frames.append(env.render(mode='rgb_array')) # 加載各個(gè)時(shí)刻圖像到幀
    show_state(env, step, info = 'CartPole_test') # 顯示渲染窗口
    action = env.action_space.sample() # 隨機(jī)動(dòng)作，需要學(xué)習(xí)的動(dòng)作模型
    # action=np.random.choice(2) # 隨機(jī)返回： 0 小車向左，1 小車向右

    observation,reward,done,info = env.step(action) # 執(zhí)行動(dòng)作并返回結(jié)果

env.close()

display_frames_as_gif(frames, SavePath = './CartPole_result.gif') # 保存運(yùn)行結(jié)果動(dòng)圖

CartPole_result.gif

# 運(yùn)行環(huán)境實(shí)例2

import gym

frames=[]
env = gym.make("LunarLander-v2")
env.reset()
env.action_space.seed(42)

observation, info = env.reset(seed=42, return_info=True)

for step in range(100):
    frames.append(env.render(mode='rgb_array')) # 加載各個(gè)時(shí)刻圖像到幀
    env.render(mode='human') # 這行不能和env定義寫在一行，否則會(huì)報(bào)錯(cuò)，原因不明
    time.sleep(0.1) # 控制顯示速度變慢
    show_state(env, step, info="LunarLander_test")
    observation, reward, done, info = env.step(env.action_space.sample())

    if done:
        observation, info = env.reset(return_info=True)

env.close()
display_frames_as_gif(frames, SavePath = './LunarLander_result.gif') # 保存運(yùn)行結(jié)果動(dòng)圖

LunarLander_result.gif